Outlook: Neuromorphic computing as an AI accelerator

Neuromorphic hardware and software for edge Al applications

Fraunhofer IIS is a world-class innovation driver in the field of neuromorphic computing. We develop chips whose architec -
ture is based on biological neural networks.

In addition, Fraunhofer IIS is working on integrated circuits for convolutional neural networks (CNNs) and spiking neural networks (SNNs) in qualified CMOS technologies that are scalable and configurable depending on the use case. Using analog circuits, our low-power, low-latency neuromorphic processor units (NPUs) for CNNs have higher parallel processing and lower on-chip data transfer requirements than von Neumann based solutions. The software tools we develop are used to train networks and to transfer the networks to the hardware.

© istock, tcareob72
Neuromorphic computing: Chips whose architecture is based on biological neural networks

Strategically important and sustainable research field 

What makes Fraunhofer IIS’s neuromorphic solutions special is that their hardware and software development is seamlessly intertwined, so that both components are coordinated from the outset. This particular hardware-software co-design flow permits the rapid development of highly complex designs, especially through our neural network (NN) synthesis generator tool.

To this end, we combine our expertise in low-power IC design, neural network (NN) algorithms, software tools, system and circuit architecture design with application-specific domain knowledge such as sensors, audio signal and image processing, message transmission and localization. By combining this expertise and domain knowledge, we can implement complete application-specific solutions that offer better system performance than generic solutions.